Five years across traditional ML, deep learning, and data science — now building Generative AI in production: RAG, agentic systems, fine-tuning, and cloud-native inference across automotive, healthcare, and e-commerce.
I've spent five years working across the AI stack — starting in traditional machine learning, deep learning, and data science, and now focused on Generative AI. A lot of my early work was deep learning on millions of real-world ECG records, and I've carried that rigor into the LLM era: RAG, agentic architectures, and fine-tuning.
What I care about is AI that actually ships. That means modular design, sensible orchestration, CI/CD, and observability — not just notebooks that work once. I've delivered AI showcased at CES, contributed to FDA-approved medical software, and built agentic automation that now runs inside enterprise workflows.
RAG pipelines and agentic systems with LangGraph, on AWS Bedrock & OpenAI.
1D-CNNs, VAEs, and signal processing on real-world ECG data.
Cloud-native deployment on AWS & Azure with Terraform, CI/CD, monitoring.
LoRA / PEFT adapters and inference tuning — leaner compute, faster pipelines.
Designing agentic Generative AI applications for large-scale e-commerce across retail, analytics, and engineering domains on AWS Bedrock and OpenAI. Shipped an agentic system that automates ServiceNow workflows for a major retailer, deployed as a Microsoft Teams bot with production-grade logging, monitoring, and access control.
Joined as a Machine Learning Engineer and grew into a Senior role over four years. Early on, end-to-end ML over millions of ECG records — Random Forest, XGBoost, and 1D-CNNs for arrhythmia detection — feeding into FDA-approved heart-failure risk software. Later led GenAI for a global automotive electronics supplier: fine-tuned LLMs with LoRA across five modules (−40% compute), improved arrhythmia accuracy 50% with a hybrid rule + ML system, cut pipeline runtime 70%, and streamlined CI/CD on Azure. Along the way — a VAE roof-change detector for an insurer (76%), a RAG HR assistant built with interns, and mentoring juniors across ML, GenAI, and RAG.
Selected projects across automotive infotainment, cardiac diagnostics, and enterprise automation.
Led Generative AI capabilities for in-vehicle infotainment for a global automotive electronics supplier — five production-ready components spanning assistant, productivity, diagnostics, on-device RAG, and entertainment.
A Python package for ECG signal analysis powering cardiac telemedicine and home sleep-testing workflows, with preprocessing and feature engineering that fed into clinical approval.
An agentic system that automates ServiceNow workflows for a major retailer, integrated as a Microsoft Teams bot and built for enterprise scale.
Led a two-person team building a predictive model for cap-approval workflows in manufacturing, reducing processing overhead and making predictions explainable to stakeholders.
Side projects and experiments — agents, automation, and the occasional computer-vision detour.
Parses a LaTeX résumé, analyzes a job description, and rewrites sections to improve role alignment and ATS relevance.
Python LocalLoraXDynamic LoRA adapters running locally on Llama 3.2 3B — hot-swapping fine-tuned behaviors at inference.
Jupyter PitStopA pit-stop for the vehicles — keeps servicing, document, and renewal dates in check so nothing lapses.
Python DropHunterA product-watch & price/stock notification bot — keeps tabs on items and pings when things change.
Python TaskPilotA lightweight task-automation helper — turning routine workflows into something hands-off.
Python OneRingToRuleThemAllOne place to rule the rest — a personal hub that ties my homelab tools and automations together.
Python HomeserverMonitoringMonitoring and alerting stack for a self-hosted Mac-mini home server running Ubuntu.
Shell ECG_Analysis_using_AI_IoTA deep-learning, IoT-enabled device for real-time heartbeat classification via ECG monitoring.
JupyterPlus a stack of computer-vision and ML experiments — a Beta-VAE on animal faces, Mask-RCNN on Rick & Morty, monocular depth (DNet), cervix-cancer classification, a Flask cardiac-disease app, and LSTM stock prediction. All on GitHub ↗
ECG Noise Classification Using Deep Learning with Feature Extraction
Springer · Journal of Signal, Image and Video ProcessingOpen to conversations about Generative AI, agentic systems, and production ML. The fastest way to reach me is email.